🧠 Knowledge Loop

Supercharge your daily reading. Turn newsletters, videos, and articles into a personalized, self-improving knowledge pipeline.

Knowledge Loop is an AI-powered personal content intelligence pipeline (designed for Antigravity). It automates daily knowledge digestion — ingesting newsletters, YouTube videos, social media posts, and web articles — then summarizing, distilling, and surfacing actionable insights through AI-powered suggestions that continuously learn and improve based on your feedback.

---

🌟 What It Does

Knowledge Loop creates a knowledge loop: you feed it content sources, it digests and summarizes them, suggests actionable next steps, learns your preferences from your feedback, and gets smarter over time.

graph LR
    Sources["📩 Newsletters\n🎬 YouTube\n🧵 Threads\n🌐 Websites"] --> Ingest["⚙️ Ingest Skills"]
    Ingest --> Reports["📄 Summary Reports"]
    Reports --> Distill["🧪 Daily Distillation"]
    Reports --> Suggest["💡 AI Suggestions"]
    Suggest --> Review["👤 User Review"]
    Review --> Prefs["🎯 Preferences"]
    Prefs --> Suggest

Loading

Core Capabilities

Skill	Description
📩 Newsletter Digest	Fetches unread newsletters from Gmail and summarizes them into Markdown reports
🎬 YouTube Transcription	Transcribes and documents YouTube videos into structured Markdown
🧵 Threads Ingestion	Extracts full content from Threads posts via browser automation
🌐 Website Ingestion	Fetches and summarizes web articles via Jina Reader API
🗞️ Daily Distillation	Synthesizes the day's reports into a Knowledge Distillation document
💡 Suggestion Review	AI-generated actionable suggestions scored by a rubric grader
📧 Gmail Management	Send, read, and manage Gmail via GWS CLI
✅ Google Tasks	Manage task lists and tasks for the Delegate workflow
🌐 Browser Automation	Automate browser tasks: navigation, scraping, screenshots

---

🚀 Getting Started

⚡ Quick Start Checklist

If you want to get up and running in 2 minutes:

1. Clone & Setup: Clone the repository and install the required third-party skills.
2. Authorize GWS: Install gws CLI and run auth commands.
3. Configure Gmail: Create Gmail label newsletter and setup filters.
4. Configure Tasks: Create a Google Tasks list named Delegate.
5. Personalize Goals: Edit your personal goals in goals.md.
6. Trigger Pipeline: Ask Antigravity: "run my daily workflow".

---

Step-by-Step Installation

1. Clone or Use Template

Click "Use this template" on GitHub, or clone locally:

git clone https://github.com/<your-username>/knowledge-loop.git
cd knowledge-loop

2. Install Node.js

Ensure Node.js (v18+) is installed. This is required to run the skills installer utility (npx skills).

3. Install Third-Party Skills

The template includes 13 built-in skills. To install the 5 third-party dependencies:

npx skills experimental_install

This installs: agent-browser, architecture-decision-records, gws-gmail, gws-shared, gws-tasks.

4. Configure GWS (Google Workspace CLI)

Required for Gmail and Google Tasks automation:

brew tap googleworkspace/cli
brew install gws
gws auth setup
gws auth login

5. Setup External Resources

The pipeline needs specific external resources to run correctly:

A. Gmail newsletter Label & Filters

The ingest-newsletter skill processes unread emails with the newsletter label (label:newsletter is:unread).

• Web UI (Recommended for auto-routing):
  1. Open Gmail.
  2. Click Create new label in the left sidebar and name it newsletter (case-sensitive).
  3. Create a search filter (e.g. from your favorite newsletter senders) and check Apply the label: newsletter (and optionally Skip the Inbox).
• CLI (Label creation only):

  gws gmail users labels create \
    --params '{"userId": "me"}' \
    --json '{"name": "newsletter", "labelListVisibility": "labelShow", "messageListVisibility": "show"}'

B. Google Tasks Delegate List

The daily-workflow skill looks for a task list named Delegate to find URLs (YouTube, Threads, Websites) to ingest.

• CLI (Recommended):

  gws tasks tasklists insert --json '{"title": "Delegate"}'

• Web UI: Open Gmail/Calendar side panel, click Tasks -> Create new list -> name it Delegate (case-sensitive).

6. Customize Your Settings

1. Edit goals.md — Define your personal goals to calibrate suggestion scoring.
2. Edit rubric_blocklist.md — Add topics you want auto-filtered from suggestions.
3. Add prompts to data/prompts/ — Fill in the template files or add your own.

---

🛠️ Usage

Trigger skills by typing natural language commands in your Antigravity chat:

Trigger phrase	Skill invoked
"summarize my newsletters"	ingest-newsletter
"run my daily workflow"	daily-workflow
"distill today's reports"	daily-distiller
"fetch this Threads post: <url>"	ingest-threads
"transcribe this YouTube video: <url>"	ingest-youtube
"summarize this page: <url>"	ingest-website
"review my suggestions"	review-suggestions

Daily Workflow Execution

The daily-workflow skill chains the entire pipeline in an optimized async/sync order:

1. Discover tasks from the Google Tasks Delegate list.
2. Fire YouTube transcriptions asynchronously in the background.
3. Process unread newsletters from Gmail.
4. Process Threads posts.
5. Process Website articles.
6. Await and complete YouTube transcription tasks.
7. Run daily knowledge distillation.
8. Review AI suggestions.

---

🏗️ Developer & Architecture Guide

Knowledge Loop is designed around a Modular, Skill-Based Agentic Architecture built explicitly for Google's Antigravity agent.

📐 System Architecture

Instead of traditional monolithic code, the agent operates directly on Skills. A skill is a self-contained bundle consisting of:

• A SKILL.md file that gives the agent instructions, system prompts, schemas, and commands.
• An optional README.md file summarizing features and usage.
• Helper references, scripts, or local testing configurations.

graph TD
    AG[Antigravity Agent] -->|Reads & Executes| Skill[".agents/skills/*"]
    Skill -->|Defines| Instructions["SKILL.md (YAML frontmatter + commands)"]
    Skill -->|Optionally Runs| CLI["External CLI Tools (gws, yt2doc, etc.)"]
    Skill -->|Reads/Writes| Data["data/ (goals, preferences, suggestions)"]

Loading

🔄 Data & Suggestion Lifecycle

sequenceDiagram
    participant U as User
    participant AG as Antigravity
    participant IS as Ingest Skills
    participant R as reports/
    participant SP as suggestions_pending.md
    participant UP as user_preferences.md

    U->>AG: "run my daily workflow"
    AG->>IS: Execute ingestion (Gmail/Tasks)
    IS->>R: Write markdown reports
    AG->>SP: Generate & score suggestions
    U->>AG: "review my suggestions"
    AG->>UP: Update learned user preferences

Loading

🗺️ Developer Knowledge Map

Refer to this map to quickly locate system files and design details:

Category	File / Path	Purpose
Agent Rules	AGENTS.md	Coding conventions, strict tool constraints, and workflow guidelines.
Glossary	CONTEXT.md	Definition of project terms like Dreamer, Trace, and Memory Injection.
Skill Logic	.agents/skills/	Source folders of all 13 built-in and third-party skills.
Preferences	user_preferences.md	Key-value store of user preferences updated via reviews.
Goal Settings	goals.md	User goals used by the suggestion scoring rubric.
Safety Hooks	prevent_dangerous_commands.sh	Hook executing before command runs to prevent unsafe operations.
Bugs & RCAs	docs/rca/	Post-mortem logs of system errors and their primary root causes.
Decision Logs	docs/decision_logs/	Session logs mapping conversational steps to architectural rationale.
Unified Backlog	backlog.md	Shared tracking document for all pending features and bug fixes.
Settings	settings.json	Global configuration for agent settings and lifecycle hooks.

---

🌐 How to Switch Language

By default, both reports and conversation default to English. You can configure them independently in user_preferences.md under the ## Configuration section:

1. Preferred Report Language: Controls the language of the generated reports and summaries (e.g., Traditional Chinese, English).
2. Preferred Conversation Language: Controls the language used by the agent when communicating directly with you in chat.

To change them:

1. Edit user_preferences.md and set the desired language values:

   ## Configuration
   - **Preferred Report Language**: Traditional Chinese
   - **Preferred Conversation Language**: Traditional Chinese

2. (Optional) Update rubric_blocklist.md — add ambiguity phrases in your target language.

---

⚙️ Maintenance & Self-Improvement

• Lessons learned are recorded in learnings/lessons.md after notable executions.
• Bugs and unexpected behaviors are documented as Root Cause Analyses in docs/rca/ (see AGENTS.md for RCA requirements).
• Architectural decisions follow the ADR format (architecture-decision-records skill).
• Suggestions build your preference profile over time via the review-suggestions skill.

---

📄 License

MIT